Illustration of the suction-diffusion-denoising process.
This is the code of pytorch version for paper: [Diffusion Suction Grasping with Large-Scale Parcel Dataset]
Illustration of the Diffusion-Suction architecture for 6DoF Pose Estimation in stacked scenarios.
Illustration of the Self-Parcel-Suction-Labeling pipeline.
Evaluation SuctionNet-1Billion dataset
Evaluation Parcel-Suction-Dataset dataset
Please clone the repository locally:
git clone https://github.com/TAO-TAO-TAO-TAO-TAO/Diffusion_Suction.git
Install the environment：
Install Pytorch. It is required that you have access to GPUs. The code is tested with Ubuntu 16.04/18.04, CUDA 10.0 and cuDNN v7.4, python3.6. Our backbone PointNet++ is borrowed from pointnet2. .Compile the CUDA layers for PointNet++, which we used in the backbone network:
cd train\Sparepart\train.py
python train.py install
Install the following Python dependencies (with pip install):
matplotlib
opencv-python
plyfile
'trimesh>=2.35.39,<2.35.40'
'networkx>=2.2,<2.3'
torch==1.1.0
torchvision==0.3.0
sklearn
h5py
nibabel
cd train\Sparepart\train.py
python train.py
Parcel-Suction-Dataset is available at here.
SuctionNet-1Billion is available at here.
Evaluation metric The python code of evaluation metric is available at here.
If you find our work useful in your research, please consider citing:
@article{huang2025diffusion,
title={Diffusion Suction Grasping with Large-Scale Parcel Dataset},
author={Huang, Ding-Tao and He, Xinyi and Hua, Debei and Yu, Dongfang and Lin, En-Te and Zeng, Long},
journal={arXiv preprint arXiv:2502.07238},
year={2025}
}
If you have any questions, please feel free to contact the authors.
Ding-Tao Huang: [email protected]